Assessment of solar utilization potential for streetlights using street view imagery and deep learning: A case study in Hong Kong

Integrating urban lighting with photovoltaics can reduce greenhouse gas emissions and support low-carbon city development. To optimize solar streetlight deployment, we propose a novel framework for detecting and geo-locating streetlights and assessing their solar utilization potential using Google Street View images and deep learning. In this framework, a hybrid method is designed to detect and locate streetlights from panoramic images; a solar energy collection estimation method is developed using panoramic images; and a solar streetlight usage potential evaluation method considering multiple factors is established. A case study in Hong Kong, China, demonstrates that: (1) the streetlight detection of our framework achieves about 90 % accuracy, with average localization accuracy within 2 m at 93 %; (2) solar energy harvested by streetlights varies significantly, from 0 to 656.78 Wh/m2/day; (3) depending on photovoltaic panel area, 7719 streetlights across three types can be replaced with solar streetlights; (4) Over 20 years, replacement can save 4.25 × 104 MWh of electricity, 1.70 × 104 t of coal, 4.24 × 104 t of CO2, providing substantial economic and environmental benefits. This work provides an effective, scalable approach to urban PV retrofitting, contributing to sustainable lighting and carbon-neutral city planning.